| 1 | //! The sole producer of [`ReadyRouteCandidate`] (#3384). |
| 2 | //! |
| 3 | //! [`RouteResolver::resolve`] is the ONLY caller of |
| 4 | //! `ReadyRouteCandidate::new`. It resolves a [`RouteRequest`] into an |
| 5 | //! executable route using: |
| 6 | //! |
| 7 | //! 1. provider from `explicit_provider` ONLY (no base-URL / prefix sniffing); |
| 8 | //! when absent, the workspace default provider scope is used. The provider |
| 9 | //! is NEVER inferred from a model prefix. |
| 10 | //! 2. the model selector, interpreted STRICTLY within that provider's scope |
| 11 | //! against resolver-provided offerings plus the provider default. The default |
| 12 | //! resolver uses [`bundled_offerings`], while tests or snapshot loaders can |
| 13 | //! inject Models.dev-derived rows. Prefixed selectors are preserved verbatim |
| 14 | //! as the [`WireModelId`]. |
| 15 | //! 3. `auto` => the [`LogicalModelRef::is_auto`] sentinel, never a literal |
| 16 | //! model. |
| 17 | //! |
| 18 | //! It encodes its OWN minimal direct/aggregator/local classification because |
| 19 | //! the tui helpers (`provider_passes_model_through` / |
| 20 | //! `accepts_custom_model_ids`) are not reachable from `crates/config`. The |
| 21 | //! classification here is deliberately NARROWER than tui's `validate_route`: |
| 22 | //! it only rejects [`RouteError::ForeignModelForDirectProvider`] for a small |
| 23 | //! set of strict direct providers given a clearly-foreign selector; |
| 24 | //! aggregators, local, and custom endpoints pass through `Ok` with |
| 25 | //! `validation.ok == true`. |
| 26 | //! |
| 27 | //! There is deliberately no prompt-text / freeform field on [`RouteRequest`], |
| 28 | //! which structurally bars prompt-content routing. |
| 29 | |
| 30 | use super::candidate::{ |
| 31 | LimitField, PricingSku, ReadyRouteCandidate, ResolvedAuthSource, ResolvedEndpoint, |
| 32 | SourcedLimitOverride, ValidationReport, |
| 33 | }; |
| 34 | use super::capabilities::{ |
| 35 | RouteCapabilities, documented_deepseek_files_api_for_route, |
| 36 | documented_moonshot_web_search_for_route, documented_zai_web_search_for_route, |
| 37 | }; |
| 38 | use super::descriptor::ProviderDescriptor; |
| 39 | use super::errors::RouteError; |
| 40 | use super::ids::{LogicalModelRef, ModelId, ProviderId, WireModelId}; |
| 41 | use super::offering::{ProviderModelOffering, RouteLimits, bundled_offerings}; |
| 42 | use crate::catalog::{CatalogOffering, bundled_catalog_offerings}; |
| 43 | use crate::provider::WirePolicy; |
| 44 | use crate::{ProviderKind, opencode_go_model_id, provider_preserves_custom_base_url_model}; |
| 45 | |
| 46 | /// A request to resolve into an executable route. |
| 47 | /// |
| 48 | /// Note the absence of any prompt-text/freeform field: the resolver cannot see |
| 49 | /// prompt content, so it cannot silently route on it. |
| 50 | #[derive(Debug, Clone, Default)] |
| 51 | pub struct RouteRequest { |
| 52 | /// Explicit provider choice. The ONLY source of provider identity. |
| 53 | pub explicit_provider: Option<ProviderKind>, |
| 54 | /// The model the caller selected (may be `auto` or prefixed). |
| 55 | pub model_selector: Option<LogicalModelRef>, |
| 56 | /// A previously-saved provider wire model id, used as scope fallback. |
| 57 | pub saved_provider_model: Option<WireModelId>, |
| 58 | /// An explicit base URL override for the endpoint. |
| 59 | pub base_url_override: Option<String>, |
| 60 | /// Sourced limit overrides, applied in order BEFORE the candidate is |
| 61 | /// constructed and recorded on it as provenance. This is the ONLY channel |
| 62 | /// for adjusting a route's effective limits: the candidate itself is |
| 63 | /// immutable once minted. |
| 64 | pub limit_overrides: Vec<SourcedLimitOverride>, |
| 65 | } |
| 66 | |
| 67 | /// Resolves [`RouteRequest`]s into [`ReadyRouteCandidate`]s. |
| 68 | #[derive(Debug, Clone)] |
| 69 | pub struct RouteResolver { |
| 70 | offerings: Vec<ProviderModelOffering>, |
| 71 | configured_offerings: Vec<(String, ProviderModelOffering)>, |
| 72 | } |
| 73 | |
| 74 | /// Offering-owned facts selected within one provider scope before the final |
| 75 | /// executable route candidate is minted. |
| 76 | struct ResolvedOffering { |
| 77 | wire_model_id: WireModelId, |
| 78 | canonical_model: Option<ModelId>, |
| 79 | endpoint_key: String, |
| 80 | limits: RouteLimits, |
| 81 | capabilities: RouteCapabilities, |
| 82 | pricing: PricingSku, |
| 83 | } |
| 84 | |
| 85 | impl ResolvedOffering { |
| 86 | fn unknown(wire_model_id: WireModelId) -> Self { |
| 87 | Self { |
| 88 | wire_model_id, |
| 89 | canonical_model: None, |
| 90 | endpoint_key: "chat".to_string(), |
| 91 | limits: RouteLimits::default(), |
| 92 | capabilities: RouteCapabilities::default(), |
| 93 | pricing: PricingSku::UnknownOrStale, |
| 94 | } |
| 95 | } |
| 96 | |
| 97 | fn from_offering(offering: &ProviderModelOffering) -> Self { |
| 98 | Self { |
| 99 | wire_model_id: offering.wire_model_id.clone(), |
| 100 | canonical_model: offering.canonical_model.clone(), |
| 101 | endpoint_key: offering.endpoint_key.clone(), |
| 102 | limits: offering.limits, |
| 103 | capabilities: offering.capabilities, |
| 104 | pricing: offering.pricing.clone(), |
| 105 | } |
| 106 | } |
| 107 | } |
| 108 | |
| 109 | impl Default for RouteResolver { |
| 110 | fn default() -> Self { |
| 111 | Self::new() |
| 112 | } |
| 113 | } |
| 114 | |
| 115 | impl RouteResolver { |
| 116 | /// Construct a resolver with CodeWhale's bundled offline offerings. |
| 117 | /// |
| 118 | /// The default offerings are the committed Models.dev-shaped catalog asset |
| 119 | /// (`crate::catalog::bundled_catalog_offerings`, real context windows and |
| 120 | /// honest per-row `cost`) merged with the tiny hand seam |
| 121 | /// ([`bundled_offerings`]). The hand seam is kept and given precedence on a |
| 122 | /// `(provider, wire id)` collision: it encodes the curated canonical-model |
| 123 | /// joins the route invariants depend on (e.g. a DeepSeek-native row and the |
| 124 | /// aggregator rows that map a prefixed wire id back to `deepseek-v4-pro`), |
| 125 | /// which generated Models.dev JSON does not prove. Asset-only rows (GLM, |
| 126 | /// Kimi, MiniMax, Qwen, …) add the real provider/model facts the picker and |
| 127 | /// candidates were previously missing. |
| 128 | #[must_use] |
| 129 | pub fn new() -> Self { |
| 130 | Self::from_offerings(default_offerings()) |
| 131 | } |
| 132 | |
| 133 | /// Construct a resolver from a provider-scoped offering catalog. |
| 134 | /// |
| 135 | /// This is the bridge for Models.dev snapshots: callers parse a catalog, |
| 136 | /// emit provider offerings, then hand those rows to the resolver without |
| 137 | /// changing route-resolution semantics. |
| 138 | #[must_use] |
| 139 | pub fn from_offerings(offerings: Vec<ProviderModelOffering>) -> Self { |
| 140 | Self { |
| 141 | offerings, |
| 142 | configured_offerings: Vec::new(), |
| 143 | } |
| 144 | } |
| 145 | |
| 146 | /// Add validated operator declarations for this exact route. This does |
| 147 | /// not grant provider-catalog authority or establish model availability. |
| 148 | pub fn with_configured_models( |
| 149 | mut self, |
| 150 | models: &[crate::catalog::configured::ConfiguredModel], |
| 151 | identity: &str, |
| 152 | provider: ProviderKind, |
| 153 | base_url: &str, |
| 154 | ) -> Self { |
| 155 | if crate::catalog::configured::validate_configured_models(models).is_err() { |
| 156 | return self; |
| 157 | } |
| 158 | for model in models |
| 159 | .iter() |
| 160 | .filter(|model| model.matches_route(identity, base_url)) |
| 161 | { |
| 162 | let mut offering = model.to_catalog_offering().to_offering(); |
| 163 | offering.provider = ProviderId::from(provider.as_str()); |
| 164 | // Preserve an adapter's established protocol for an existing ID. |
| 165 | // The label/config declaration cannot choose a new wire dialect. |
| 166 | if let Some(existing) = self.offerings.iter().find(|row| { |
| 167 | row.provider == offering.provider && row.wire_model_id == offering.wire_model_id |
| 168 | }) { |
| 169 | offering.endpoint_key.clone_from(&existing.endpoint_key); |
| 170 | offering.default_for_provider = existing.default_for_provider; |
| 171 | } else if ProviderDescriptor::for_kind(provider).wire_policy() == WirePolicy::ModelAware |
| 172 | && provider != ProviderKind::Deepseek |
| 173 | { |
| 174 | continue; |
| 175 | } |
| 176 | // Positive declarations are not verified executable capabilities. |
| 177 | // Explicit negatives can still restrict a route conservatively. |
| 178 | let declared = offering.capabilities; |
| 179 | offering.capabilities = RouteCapabilities { |
| 180 | attachments: if model.attachment == Some(false) { |
| 181 | declared.attachments |
| 182 | } else { |
| 183 | Default::default() |
| 184 | }, |
| 185 | image_input: if declared.image_input == super::CapabilityState::Unsupported { |
| 186 | declared.image_input |
| 187 | } else { |
| 188 | Default::default() |
| 189 | }, |
| 190 | reasoning: if model.reasoning == Some(false) { |
| 191 | declared.reasoning |
| 192 | } else { |
| 193 | Default::default() |
| 194 | }, |
| 195 | native_tool_calls: if model.tool_call == Some(false) { |
| 196 | declared.native_tool_calls |
| 197 | } else { |
| 198 | Default::default() |
| 199 | }, |
| 200 | structured_output: if model.structured_output == Some(false) { |
| 201 | declared.structured_output |
| 202 | } else { |
| 203 | Default::default() |
| 204 | }, |
| 205 | ..RouteCapabilities::default() |
| 206 | }; |
| 207 | self.configured_offerings |
| 208 | .push((crate::catalog::base_url_fingerprint(base_url), offering)); |
| 209 | } |
| 210 | self |
| 211 | } |
| 212 | |
| 213 | /// Resolve a request into an executable route candidate. |
| 214 | /// |
| 215 | /// # Errors |
| 216 | /// Returns [`RouteError`] when the model is empty, the provider is invalid, |
| 217 | /// or a clearly-foreign model is requested for a strict direct provider. |
| 218 | pub fn resolve(&self, req: &RouteRequest) -> Result<ReadyRouteCandidate, RouteError> { |
| 219 | self.resolve_inner(req, false) |
| 220 | } |
| 221 | |
| 222 | /// Resolve with catalog facts authenticated against this exact endpoint. |
| 223 | /// |
| 224 | /// The ordinary [`Self::resolve`] path strips capabilities and pricing when |
| 225 | /// a route uses a custom base URL, because a same-named first-party model is |
| 226 | /// not evidence about an arbitrary proxy. Callers may use this seam only |
| 227 | /// when the injected offering came from the selected provider identity's |
| 228 | /// own endpoint and its base-URL fingerprint matches the request endpoint. |
| 229 | /// All normal routing and protocol validation still applies. |
| 230 | pub fn resolve_with_endpoint_catalog_authority( |
| 231 | &self, |
| 232 | req: &RouteRequest, |
| 233 | ) -> Result<ReadyRouteCandidate, RouteError> { |
| 234 | self.resolve_inner(req, true) |
| 235 | } |
| 236 | |
| 237 | fn resolve_inner( |
| 238 | &self, |
| 239 | req: &RouteRequest, |
| 240 | endpoint_catalog_authoritative: bool, |
| 241 | ) -> Result<ReadyRouteCandidate, RouteError> { |
| 242 | if self.configured_offerings.is_empty() { |
| 243 | return self.resolve_scoped(req, endpoint_catalog_authoritative); |
| 244 | } |
| 245 | let mut scoped = self.clone(); |
| 246 | scoped.configured_offerings.clear(); |
| 247 | let provider = req.explicit_provider.unwrap_or_default(); |
| 248 | let base_url = req |
| 249 | .base_url_override |
| 250 | .as_deref() |
| 251 | .unwrap_or_else(|| ProviderDescriptor::for_kind(provider).default_base_url()); |
| 252 | let fingerprint = crate::catalog::base_url_fingerprint(base_url); |
| 253 | let selected_id = req |
| 254 | .model_selector |
| 255 | .as_ref() |
| 256 | .map(LogicalModelRef::raw) |
| 257 | .or_else(|| req.saved_provider_model.as_ref().map(WireModelId::as_str)); |
| 258 | for (endpoint, offering) in &self.configured_offerings { |
| 259 | if !base_url.contains(['@', '?', '#']) |
| 260 | && *endpoint == fingerprint |
| 261 | && offering.provider.as_str() == provider.as_str() |
| 262 | && selected_id == Some(offering.wire_model_id.as_str()) |
| 263 | { |
| 264 | scoped.offerings.retain(|row| { |
| 265 | row.provider != offering.provider || row.wire_model_id != offering.wire_model_id |
| 266 | }); |
| 267 | scoped.offerings.push(offering.clone()); |
| 268 | scoped |
| 269 | .configured_offerings |
| 270 | .push((endpoint.clone(), offering.clone())); |
| 271 | } |
| 272 | } |
| 273 | scoped.resolve_scoped(req, endpoint_catalog_authoritative) |
| 274 | } |
| 275 | |
| 276 | fn resolve_scoped( |
| 277 | &self, |
| 278 | req: &RouteRequest, |
| 279 | endpoint_catalog_authoritative: bool, |
| 280 | ) -> Result<ReadyRouteCandidate, RouteError> { |
| 281 | // 1. Provider scope from explicit choice only; default otherwise. |
| 282 | // The provider is NEVER inferred from a model prefix. |
| 283 | let provider_kind = req.explicit_provider.unwrap_or_default(); |
| 284 | if provider_kind == ProviderKind::Antigravity { |
| 285 | return Err(RouteError::InvalidProvider( |
| 286 | crate::LEGACY_ANTIGRAVITY_TOMBSTONE_MESSAGE.to_string(), |
| 287 | )); |
| 288 | } |
| 289 | let descriptor = ProviderDescriptor::for_kind(provider_kind); |
| 290 | let provider_id = descriptor.id(); |
| 291 | let default_offering = self.default_offering(&provider_id); |
| 292 | |
| 293 | // 2. Determine the logical selector from explicit choice, then the |
| 294 | // saved-model fallback, then the provider default. |
| 295 | let logical_model = match &req.model_selector { |
| 296 | Some(selector) => selector.clone(), |
| 297 | None => { |
| 298 | // No selector: fall back to saved wire model, then provider |
| 299 | // default. Both stay in the resolved provider's scope. |
| 300 | let raw = req |
| 301 | .saved_provider_model |
| 302 | .as_ref() |
| 303 | .map(|w| w.as_str().to_string()) |
| 304 | .unwrap_or_else(|| { |
| 305 | default_offering.map_or_else( |
| 306 | || descriptor.default_wire_model().as_str().to_string(), |
| 307 | |offering| offering.wire_model_id.as_str().to_string(), |
| 308 | ) |
| 309 | }); |
| 310 | LogicalModelRef::from(raw) |
| 311 | } |
| 312 | }; |
| 313 | |
| 314 | // Reject an empty selector from ANY source (explicit, saved, or a |
| 315 | // degenerate default), not just an empty explicit selector. |
| 316 | if logical_model.raw().is_empty() { |
| 317 | return Err(RouteError::EmptyModel); |
| 318 | } |
| 319 | |
| 320 | // 3. `auto` is an opt-in sentinel: resolve to the provider default wire |
| 321 | // id without treating "auto" as a literal model name. |
| 322 | let is_auto = logical_model.is_auto(); |
| 323 | |
| 324 | // 4. Map the selector to a wire id within provider scope. |
| 325 | // Prefixed selectors are preserved VERBATIM as the wire id. |
| 326 | let custom_endpoint = |
| 327 | request_uses_custom_endpoint(&descriptor, req.base_url_override.as_deref()); |
| 328 | let class = if custom_endpoint { |
| 329 | ProviderClass::LocalOrCustom |
| 330 | } else { |
| 331 | classify(provider_kind) |
| 332 | }; |
| 333 | let model_aware = descriptor.wire_policy() == WirePolicy::ModelAware; |
| 334 | // A model-aware protocol row is an exact provider-endpoint fact. |
| 335 | // OpenCode Zen's published roster is closed; DeepSeek's direct route |
| 336 | // deliberately preserves its existing future-model pass-through and |
| 337 | // sends unknown bare ids over Chat until an exact Responses row exists. |
| 338 | // A custom DeepSeek-compatible endpoint retains that pass-through, but |
| 339 | // a custom URL must not weaken another model-aware provider's closed |
| 340 | // protocol roster. |
| 341 | let require_catalog_match = model_aware && provider_kind != ProviderKind::Deepseek; |
| 342 | let mut selected = if is_auto { |
| 343 | match default_offering { |
| 344 | None if require_catalog_match => { |
| 345 | return Err(RouteError::UnsupportedModelProtocol { |
| 346 | provider: provider_id.clone(), |
| 347 | model: descriptor.default_wire_model().as_str().to_string(), |
| 348 | endpoint_key: "unproven".to_string(), |
| 349 | }); |
| 350 | } |
| 351 | None => ResolvedOffering::unknown(descriptor.default_wire_model()), |
| 352 | Some(offering) => ResolvedOffering::from_offering(offering), |
| 353 | } |
| 354 | } else { |
| 355 | self.scope_selector( |
| 356 | provider_kind, |
| 357 | &provider_id, |
| 358 | &logical_model, |
| 359 | class, |
| 360 | require_catalog_match, |
| 361 | )? |
| 362 | }; |
| 363 | if provider_kind == ProviderKind::OpencodeGo { |
| 364 | selected.endpoint_key = |
| 365 | crate::opencode_go_endpoint_key(selected.wire_model_id.as_str()) |
| 366 | .ok_or_else(|| RouteError::UnsupportedModelProtocol { |
| 367 | provider: provider_id.clone(), |
| 368 | model: selected.wire_model_id.as_str().to_string(), |
| 369 | endpoint_key: "unproven".to_string(), |
| 370 | })? |
| 371 | .to_string(); |
| 372 | } |
| 373 | if provider_kind == ProviderKind::Deepseek && custom_endpoint { |
| 374 | selected.endpoint_key = "chat".to_string(); |
| 375 | } |
| 376 | if custom_endpoint && !endpoint_catalog_authoritative { |
| 377 | // Capabilities and pricing belong to the exact provider endpoint |
| 378 | // offering that reported them. Reusing a provider enum and a |
| 379 | // first-party model id against a custom compatible endpoint does |
| 380 | // not prove that proxy serves the same canonical model, limits, |
| 381 | // modality, tool, reasoning, or billing contract. Keep the |
| 382 | // caller's wire model id, but clear every unowned offering fact at |
| 383 | // the authority boundary instead of presenting it as verified. |
| 384 | // The endpoint_key/protocol stays: it is the provider adapter's |
| 385 | // wire contract (a model-aware roster row or fixed policy), not an |
| 386 | // endpoint-catalog fact, and coercing it to Chat would silently |
| 387 | // change how a Responses- or Messages-bound route speaks. |
| 388 | // Deepseek's custom-endpoint Chat pass-through is handled above. |
| 389 | selected.canonical_model = None; |
| 390 | selected.limits = RouteLimits::default(); |
| 391 | selected.capabilities = RouteCapabilities::default(); |
| 392 | selected.pricing = PricingSku::UnknownOrStale; |
| 393 | } |
| 394 | let base_url = req |
| 395 | .base_url_override |
| 396 | .as_deref() |
| 397 | .unwrap_or_else(|| descriptor.default_base_url()); |
| 398 | let mut declared_limit_overrides = Vec::new(); |
| 399 | if let Some((_, offering)) = |
| 400 | self.configured_offerings |
| 401 | .iter() |
| 402 | .find(|(fingerprint, offering)| { |
| 403 | !base_url.contains(['@', '?', '#']) |
| 404 | && offering.provider == provider_id |
| 405 | && offering.wire_model_id == selected.wire_model_id |
| 406 | && req |
| 407 | .model_selector |
| 408 | .as_ref() |
| 409 | .map(LogicalModelRef::raw) |
| 410 | .or_else(|| req.saved_provider_model.as_ref().map(WireModelId::as_str)) |
| 411 | == Some(offering.wire_model_id.as_str()) |
| 412 | && *fingerprint == crate::catalog::base_url_fingerprint(base_url) |
| 413 | }) |
| 414 | { |
| 415 | selected.canonical_model = None; |
| 416 | selected.limits = offering.limits; |
| 417 | selected.capabilities = offering.capabilities; |
| 418 | selected.pricing = offering.pricing.clone(); |
| 419 | for (field, value) in [ |
| 420 | (LimitField::ContextTokens, offering.limits.context_tokens), |
| 421 | (LimitField::InputTokens, offering.limits.input_tokens), |
| 422 | (LimitField::OutputTokens, offering.limits.output_tokens), |
| 423 | ] { |
| 424 | declared_limit_overrides.push(SourcedLimitOverride { |
| 425 | field, |
| 426 | value, |
| 427 | source: super::OverrideSource::UserModelMetadata, |
| 428 | }); |
| 429 | } |
| 430 | } |
| 431 | if provider_kind == ProviderKind::Zai { |
| 432 | let effective_base_url = req |
| 433 | .base_url_override |
| 434 | .as_deref() |
| 435 | .unwrap_or_else(|| descriptor.default_base_url()); |
| 436 | selected.capabilities.server_side_web_search = documented_zai_web_search_for_route( |
| 437 | provider_kind, |
| 438 | selected.wire_model_id.as_str(), |
| 439 | effective_base_url, |
| 440 | ); |
| 441 | } |
| 442 | if provider_kind == ProviderKind::Moonshot { |
| 443 | let effective_base_url = req |
| 444 | .base_url_override |
| 445 | .as_deref() |
| 446 | .unwrap_or_else(|| descriptor.default_base_url()); |
| 447 | selected.capabilities.server_side_web_search = documented_moonshot_web_search_for_route( |
| 448 | provider_kind, |
| 449 | selected.wire_model_id.as_str(), |
| 450 | effective_base_url, |
| 451 | ); |
| 452 | } |
| 453 | if matches!( |
| 454 | provider_kind, |
| 455 | ProviderKind::Deepseek | ProviderKind::DeepseekAnthropic |
| 456 | ) { |
| 457 | let effective_base_url = req |
| 458 | .base_url_override |
| 459 | .as_deref() |
| 460 | .unwrap_or_else(|| descriptor.default_base_url()); |
| 461 | selected.capabilities.files_api = documented_deepseek_files_api_for_route( |
| 462 | provider_kind, |
| 463 | selected.wire_model_id.as_str(), |
| 464 | effective_base_url, |
| 465 | ); |
| 466 | } |
| 467 | |
| 468 | let protocol = descriptor |
| 469 | .protocol_for_endpoint(&selected.endpoint_key) |
| 470 | .ok_or_else(|| RouteError::UnsupportedModelProtocol { |
| 471 | provider: provider_id.clone(), |
| 472 | model: selected.wire_model_id.as_str().to_string(), |
| 473 | endpoint_key: selected.endpoint_key.clone(), |
| 474 | })?; |
| 475 | let endpoint = ResolvedEndpoint { |
| 476 | base_url: req |
| 477 | .base_url_override |
| 478 | .clone() |
| 479 | .unwrap_or_else(|| descriptor.default_base_url().to_string()), |
| 480 | endpoint_key: selected.endpoint_key, |
| 481 | protocol, |
| 482 | }; |
| 483 | |
| 484 | // Advisory validation (#1519): a non-loopback `http://` endpoint sends |
| 485 | // credentials in plaintext. This is advisory, not a hard fail, so |
| 486 | // `ok` stays true and local `http://localhost` runtimes (Ollama / vLLM / |
| 487 | // SGLang defaults) stay clean. |
| 488 | let mut messages = Vec::new(); |
| 489 | if endpoint_uses_insecure_http(&endpoint.base_url) { |
| 490 | messages |
| 491 | .push("endpoint uses insecure http:// (credentials sent in plaintext)".to_string()); |
| 492 | } |
| 493 | let validation = ValidationReport { ok: true, messages }; |
| 494 | |
| 495 | // Apply caller-requested limit overrides in order, BEFORE the candidate |
| 496 | // is minted. The candidate is immutable afterwards; the applied |
| 497 | // overrides are recorded on it as provenance. |
| 498 | let mut limits = selected.limits; |
| 499 | for limit_override in &req.limit_overrides { |
| 500 | match limit_override.field { |
| 501 | LimitField::ContextTokens => limits.context_tokens = limit_override.value, |
| 502 | LimitField::InputTokens => limits.input_tokens = limit_override.value, |
| 503 | LimitField::OutputTokens => limits.output_tokens = limit_override.value, |
| 504 | } |
| 505 | } |
| 506 | |
| 507 | Ok(ReadyRouteCandidate::new( |
| 508 | provider_id, |
| 509 | provider_kind, |
| 510 | logical_model, |
| 511 | selected.canonical_model, |
| 512 | selected.wire_model_id, |
| 513 | endpoint, |
| 514 | // The resolver never inspects credentials: auth is honestly |
| 515 | // `Unresolved` at resolution time, not a claimed `Missing`. |
| 516 | ResolvedAuthSource::Unresolved, |
| 517 | protocol, |
| 518 | limits, |
| 519 | selected.capabilities, |
| 520 | // #3085: honest pricing projected from the matched offering (the |
| 521 | // catalog layer maps sourced cost → SKU); `UnknownOrStale` whenever |
| 522 | // no offering was matched or the offering carried no price. |
| 523 | Some(selected.pricing), |
| 524 | validation, |
| 525 | declared_limit_overrides |
| 526 | .into_iter() |
| 527 | .chain(req.limit_overrides.iter().cloned()) |
| 528 | .collect(), |
| 529 | )) |
| 530 | } |
| 531 | |
| 532 | /// Interpret a concrete (non-auto) selector strictly within provider scope. |
| 533 | fn scope_selector( |
| 534 | &self, |
| 535 | provider_kind: ProviderKind, |
| 536 | provider_id: &ProviderId, |
| 537 | logical_model: &LogicalModelRef, |
| 538 | class: ProviderClass, |
| 539 | require_catalog_match: bool, |
| 540 | ) -> Result<ResolvedOffering, RouteError> { |
| 541 | // Go's provider roster owns each model's protocol, including when a |
| 542 | // custom base URL is configured. Unknown IDs must not fall through. |
| 543 | let raw = if provider_kind == ProviderKind::OpencodeGo { |
| 544 | opencode_go_model_id(logical_model.raw()).ok_or_else(|| { |
| 545 | RouteError::UnsupportedModelProtocol { |
| 546 | provider: provider_id.clone(), |
| 547 | model: logical_model.raw().to_string(), |
| 548 | endpoint_key: "unproven".to_string(), |
| 549 | } |
| 550 | })? |
| 551 | } else if provider_kind == ProviderKind::OpencodeZen { |
| 552 | logical_model |
| 553 | .raw() |
| 554 | .strip_prefix("opencode/") |
| 555 | .or_else(|| logical_model.raw().strip_prefix("opencode-zen/")) |
| 556 | .unwrap_or_else(|| logical_model.raw()) |
| 557 | } else if self.configured_offerings.iter().any(|(_, offering)| { |
| 558 | offering.provider == *provider_id |
| 559 | && offering.wire_model_id.as_str() == logical_model.raw() |
| 560 | }) { |
| 561 | // An explicitly declared wire ID is not a convenience selector. |
| 562 | logical_model.raw() |
| 563 | } else if provider_kind == ProviderKind::Concentrate { |
| 564 | // Concentrate's own namespace is not part of its wire ids. |
| 565 | // `concentrate/auto` is the explicit spelling for the gateway's |
| 566 | // `auto` router — Codewhale's bare `auto` is the resolver sentinel |
| 567 | // above, never a literal model id — and `concentrate/<id>` is |
| 568 | // `<id>`. Upstream prefixes (`openai/gpt-5.6-sol`) stay verbatim: |
| 569 | // they pin the upstream provider inside the gateway. |
| 570 | logical_model |
| 571 | .raw() |
| 572 | .strip_prefix("concentrate/") |
| 573 | .unwrap_or_else(|| logical_model.raw()) |
| 574 | } else { |
| 575 | provider_scoped_wire_alias(provider_kind, logical_model.raw(), class) |
| 576 | }; |
| 577 | |
| 578 | // This list was scoped to the exact endpoint and raw selector in |
| 579 | // resolve_inner, after closed-protocol admission in the builder. |
| 580 | // Prefer that literal declaration over a bundled canonical alias. |
| 581 | // Keep this after provider protocol/allowlist normalization above; |
| 582 | // a declaration cannot preserve an alias those guards must rewrite. |
| 583 | if let Some((_, offering)) = self.configured_offerings.iter().find(|(_, offering)| { |
| 584 | offering.provider == *provider_id && offering.wire_model_id.as_str() == raw |
| 585 | }) { |
| 586 | return Ok(ResolvedOffering::from_offering(offering)); |
| 587 | } |
| 588 | |
| 589 | // Try to match a catalog offering owned by THIS provider, either by |
| 590 | // canonical model id or by exact wire id. This keeps interpretation |
| 591 | // inside provider scope; offerings from other providers are ignored. |
| 592 | // DeepSeek and Z.ai also publish marketing-cased wire ids while saved |
| 593 | // selectors can be lowercase. Defer that fallback until exact matching |
| 594 | // is exhausted, and only accept a unique provider-owned match so |
| 595 | // catalog order can never choose between case-distinct model ids. |
| 596 | let allow_casefold_wire_match = class == ProviderClass::StrictDirect |
| 597 | && matches!(provider_kind, ProviderKind::Deepseek | ProviderKind::Zai); |
| 598 | let mut casefold_match = None; |
| 599 | let mut casefold_ambiguous = false; |
| 600 | for offering in &self.offerings { |
| 601 | if offering.provider != *provider_id { |
| 602 | continue; |
| 603 | } |
| 604 | let matches_canonical = offering |
| 605 | .canonical_model |
| 606 | .as_ref() |
| 607 | .is_some_and(|m| m.as_str() == raw); |
| 608 | let matches_wire = offering.wire_model_id.as_str() == raw; |
| 609 | if matches_canonical || matches_wire { |
| 610 | return Ok(ResolvedOffering::from_offering(offering)); |
| 611 | } |
| 612 | if allow_casefold_wire_match |
| 613 | && offering.wire_model_id.as_str().eq_ignore_ascii_case(raw) |
| 614 | { |
| 615 | if casefold_match.is_some() { |
| 616 | casefold_ambiguous = true; |
| 617 | } else { |
| 618 | casefold_match = Some(offering); |
| 619 | } |
| 620 | } |
| 621 | } |
| 622 | if !casefold_ambiguous && let Some(offering) = casefold_match { |
| 623 | return Ok(ResolvedOffering::from_offering(offering)); |
| 624 | } |
| 625 | |
| 626 | // No catalog match. Apply class-specific pass-through rules. |
| 627 | match class { |
| 628 | ProviderClass::StrictDirect => { |
| 629 | if self.selector_matches_other_provider_offering(provider_id, raw) { |
| 630 | return Err(RouteError::ForeignModelForDirectProvider { |
| 631 | provider: provider_id.clone(), |
| 632 | model: raw.to_string(), |
| 633 | }); |
| 634 | } |
| 635 | // A clearly-foreign selector for a strict direct provider is |
| 636 | // rejected. "Clearly foreign" = it carries an aggregator/org |
| 637 | // namespace prefix, which a direct provider never expects. |
| 638 | if logical_model.namespace_hint().is_some() { |
| 639 | return Err(RouteError::ForeignModelForDirectProvider { |
| 640 | provider: provider_id.clone(), |
| 641 | model: raw.to_string(), |
| 642 | }); |
| 643 | } |
| 644 | if require_catalog_match { |
| 645 | return Err(RouteError::UnsupportedModelProtocol { |
| 646 | provider: provider_id.clone(), |
| 647 | model: raw.to_string(), |
| 648 | endpoint_key: "unproven".to_string(), |
| 649 | }); |
| 650 | } |
| 651 | // A bare, unknown model on a strict direct provider is passed |
| 652 | // through verbatim (the provider validates it server-side). No |
| 653 | // offering matched, so pricing is honestly unknown (#3085). |
| 654 | Ok(ResolvedOffering::unknown(WireModelId::from(raw))) |
| 655 | } |
| 656 | // Aggregators, local runtimes, and custom OpenAI-compatible |
| 657 | // endpoints legitimately accept arbitrary / prefixed ids verbatim. |
| 658 | ProviderClass::Aggregator | ProviderClass::LocalOrCustom => { |
| 659 | if require_catalog_match { |
| 660 | // The Codewhale API's protocol roster is the *account's* |
| 661 | // live catalog, not a compiled list: a customer who |
| 662 | // connects a new provider gets new `provider/model` rows |
| 663 | // without a Codewhale release. Failing closed on a row the |
| 664 | // local catalog has not seen yet would make the account's |
| 665 | // own connected provider unreachable, so infer the |
| 666 | // protocol from the namespace the account API itself uses |
| 667 | // (`anthropic/` is the Messages passthrough; everything |
| 668 | // else is Chat Completions). |
| 669 | if provider_kind == ProviderKind::Codewhale { |
| 670 | return Ok(ResolvedOffering { |
| 671 | wire_model_id: WireModelId::from(raw), |
| 672 | canonical_model: None, |
| 673 | endpoint_key: super::codewhale_endpoint_key_for_model(raw).to_string(), |
| 674 | limits: RouteLimits::default(), |
| 675 | capabilities: RouteCapabilities::default(), |
| 676 | pricing: PricingSku::UnknownOrStale, |
| 677 | }); |
| 678 | } |
| 679 | // Opencode Zen serves Muse Spark exclusively over Responses. |
| 680 | // Handle any future muse-spark variant (e.g. -free suffix) |
| 681 | // even when no exact bundled offering exists — fail open to |
| 682 | // responses rather than failing closed to "unproven". |
| 683 | if provider_kind == ProviderKind::OpencodeZen |
| 684 | && raw.to_ascii_lowercase().contains("muse-spark") |
| 685 | { |
| 686 | return Ok(ResolvedOffering { |
| 687 | wire_model_id: WireModelId::from(raw), |
| 688 | canonical_model: None, |
| 689 | endpoint_key: "responses".to_string(), |
| 690 | limits: RouteLimits::default(), |
| 691 | capabilities: RouteCapabilities::default(), |
| 692 | pricing: PricingSku::UnknownOrStale, |
| 693 | }); |
| 694 | } |
| 695 | return Err(RouteError::UnsupportedModelProtocol { |
| 696 | provider: provider_id.clone(), |
| 697 | model: raw.to_string(), |
| 698 | endpoint_key: "unproven".to_string(), |
| 699 | }); |
| 700 | } |
| 701 | // No offering matched: pricing is honestly unknown (#3085). |
| 702 | Ok(ResolvedOffering::unknown(WireModelId::from(raw))) |
| 703 | } |
| 704 | } |
| 705 | } |
| 706 | |
| 707 | fn default_offering(&self, provider_id: &ProviderId) -> Option<&ProviderModelOffering> { |
| 708 | self.offerings |
| 709 | .iter() |
| 710 | .find(|offering| offering.provider == *provider_id && offering.default_for_provider) |
| 711 | } |
| 712 | |
| 713 | /// True when `raw` names an offering that lives on a *different* provider. |
| 714 | /// |
| 715 | /// The `wire_model_id` arm catches the common case (a bare id another |
| 716 | /// provider serves). The `canonical_model` arm covers catalog rows whose |
| 717 | /// canonical id is slash-free: Models.dev canonical ids normally contain a |
| 718 | /// namespace (`zhipuai/glm-5.2`) and are already caught by the |
| 719 | /// `namespace_hint()` guard at the call site, but a bare canonical id (or a |
| 720 | /// hand-authored offering) would slip through wire-id matching alone. It is |
| 721 | /// kept deliberately so a bare canonical selector cannot masquerade as a |
| 722 | /// pass-through model on the wrong provider. |
| 723 | fn selector_matches_other_provider_offering( |
| 724 | &self, |
| 725 | provider_id: &ProviderId, |
| 726 | raw: &str, |
| 727 | ) -> bool { |
| 728 | self.offerings.iter().any(|offering| { |
| 729 | offering.provider != *provider_id |
| 730 | && (offering.wire_model_id.as_str() == raw |
| 731 | || offering |
| 732 | .canonical_model |
| 733 | .as_ref() |
| 734 | .is_some_and(|model| model.as_str() == raw)) |
| 735 | }) |
| 736 | } |
| 737 | } |
| 738 | |
| 739 | /// Normalize aliases whose provider wire identity is publicly documented but |
| 740 | /// intentionally absent from the offline offering catalog. Keeping this seam |
| 741 | /// provider-scoped avoids claiming unverified limits or pricing while ensuring |
| 742 | /// receipts and HTTP requests carry the exact upstream model id. |
| 743 | fn provider_scoped_wire_alias( |
| 744 | provider_kind: ProviderKind, |
| 745 | raw: &str, |
| 746 | class: ProviderClass, |
| 747 | ) -> &str { |
| 748 | if class != ProviderClass::LocalOrCustom { |
| 749 | if provider_kind == ProviderKind::Together |
| 750 | && (raw.eq_ignore_ascii_case("inkling") || raw.eq_ignore_ascii_case("together-inkling")) |
| 751 | { |
| 752 | return "thinkingmachines/inkling"; |
| 753 | } |
| 754 | if provider_kind == ProviderKind::Openrouter |
| 755 | && (raw.eq_ignore_ascii_case("qwen3.7-plus") |
| 756 | || raw.eq_ignore_ascii_case("qwen-3.7-plus")) |
| 757 | { |
| 758 | return "qwen/qwen3.7-plus"; |
| 759 | } |
| 760 | } |
| 761 | raw |
| 762 | } |
| 763 | |
| 764 | /// Build the default resolver offerings from the bundled Models.dev asset. |
| 765 | /// |
| 766 | /// Curated transport rows win a `(provider, wire id)` collision over the asset; |
| 767 | /// all other offerings continue to come from Models.dev. |
| 768 | fn default_offerings() -> Vec<ProviderModelOffering> { |
| 769 | let mut seen: std::collections::HashSet<(String, String)> = std::collections::HashSet::new(); |
| 770 | let mut out = Vec::new(); |
| 771 | let asset_rows = bundled_catalog_offerings() |
| 772 | .iter() |
| 773 | .map(CatalogOffering::to_offering) |
| 774 | .collect::<Vec<_>>(); |
| 775 | // Seam first so it wins identity collisions, then asset-only rows follow. |
| 776 | let go_metadata: std::collections::HashMap<_, _> = asset_rows |
| 777 | .iter() |
| 778 | .filter(|row| row.provider.as_str() == "opencode-go") |
| 779 | .map(|row| (row.wire_model_id.as_str(), row)) |
| 780 | .collect(); |
| 781 | for mut offering in bundled_offerings() |
| 782 | .into_iter() |
| 783 | .chain(asset_rows.iter().cloned()) |
| 784 | { |
| 785 | // The transport seam must not erase already sourced Go limits/prices. |
| 786 | if offering.provider.as_str() == "opencode-go" |
| 787 | && let Some(metadata) = go_metadata.get(offering.wire_model_id.as_str()) |
| 788 | { |
| 789 | offering |
| 790 | .canonical_model |
| 791 | .clone_from(&metadata.canonical_model); |
| 792 | offering.limits = metadata.limits; |
| 793 | offering.capabilities = metadata.capabilities; |
| 794 | offering.pricing.clone_from(&metadata.pricing); |
| 795 | } |
| 796 | let key = ( |
| 797 | offering.provider.as_str().to_string(), |
| 798 | offering.wire_model_id.as_str().to_string(), |
| 799 | ); |
| 800 | if seen.insert(key) { |
| 801 | out.push(offering); |
| 802 | } |
| 803 | } |
| 804 | out |
| 805 | } |
| 806 | |
| 807 | /// The resolver's minimal route classification. |
| 808 | /// |
| 809 | /// Intentionally narrower than tui's `validate_route`. |
| 810 | #[derive(Debug, Clone, Copy, PartialEq, Eq)] |
| 811 | enum ProviderClass { |
| 812 | /// Strict direct provider: rejects clearly-foreign (prefixed) selectors. |
| 813 | StrictDirect, |
| 814 | /// Aggregator: serves many catalogs under prefixed wire ids. |
| 815 | Aggregator, |
| 816 | /// Local runtime or custom OpenAI-compatible endpoint: pass-through. |
| 817 | LocalOrCustom, |
| 818 | } |
| 819 | |
| 820 | /// Classify a provider kind for resolver pass-through rules. |
| 821 | /// |
| 822 | /// Only a SMALL set of providers are strict-direct. Everything else passes |
| 823 | /// through, so the resolver stays permissive by default. |
| 824 | fn classify(kind: ProviderKind) -> ProviderClass { |
| 825 | match kind { |
| 826 | // Strict first-party direct providers. |
| 827 | ProviderKind::Deepseek | ProviderKind::Zai => ProviderClass::StrictDirect, |
| 828 | // Local runtimes / custom OpenAI-compatible endpoints. |
| 829 | ProviderKind::Ollama | ProviderKind::Vllm | ProviderKind::Sglang | ProviderKind::Openai => { |
| 830 | ProviderClass::LocalOrCustom |
| 831 | } |
| 832 | // Everything else is treated as an aggregator-style pass-through. |
| 833 | _ => ProviderClass::Aggregator, |
| 834 | } |
| 835 | } |
| 836 | |
| 837 | fn request_uses_custom_endpoint( |
| 838 | descriptor: &ProviderDescriptor, |
| 839 | base_url_override: Option<&str>, |
| 840 | ) -> bool { |
| 841 | base_url_override |
| 842 | .is_some_and(|base_url| provider_preserves_custom_base_url_model(descriptor.kind, base_url)) |
| 843 | } |
| 844 | |
| 845 | /// True when `base_url` is an `http://` endpoint whose host is NOT loopback |
| 846 | /// (#1519). Such an endpoint sends credentials in plaintext over the network; |
| 847 | /// loopback (`localhost` / `127.0.0.1` / `::1`) is exempt because local |
| 848 | /// runtimes (Ollama / vLLM / SGLang) default to plain `http://localhost`. |
| 849 | fn endpoint_uses_insecure_http(base_url: &str) -> bool { |
| 850 | let trimmed = base_url.trim(); |
| 851 | // Scheme match is case-insensitive but must be `http`, not `https`. |
| 852 | let Some(rest) = strip_http_scheme(trimmed) else { |
| 853 | return false; |
| 854 | }; |
| 855 | !is_loopback_host(host_of_authority(rest)) |
| 856 | } |
| 857 | |
| 858 | /// Strip a leading case-insensitive `http://` scheme, returning the remainder. |
| 859 | /// Returns `None` for any other scheme (including `https://`) or no scheme. |
| 860 | fn strip_http_scheme(base_url: &str) -> Option<&str> { |
| 861 | let idx = base_url.find("://")?; |
| 862 | let (scheme, rest) = base_url.split_at(idx); |
| 863 | if scheme.eq_ignore_ascii_case("http") { |
| 864 | Some(&rest[3..]) |
| 865 | } else { |
| 866 | None |
| 867 | } |
| 868 | } |
| 869 | |
| 870 | /// Extract the bare host from an authority+path string: take the authority up |
| 871 | /// to the first `/`, drop any `user@` userinfo and `:port` suffix, and unwrap |
| 872 | /// `[..]` IPv6 brackets. |
| 873 | fn host_of_authority(rest: &str) -> &str { |
| 874 | let authority = rest.split('/').next().unwrap_or(rest); |
| 875 | // Drop userinfo (`user:pass@host`) if present. |
| 876 | let authority = authority.rsplit('@').next().unwrap_or(authority); |
| 877 | if let Some(inner) = authority.strip_prefix('[') { |
| 878 | // Bracketed IPv6 literal: host is everything up to the closing bracket. |
| 879 | return inner.split(']').next().unwrap_or(inner); |
| 880 | } |
| 881 | // Otherwise strip a trailing `:port`. |
| 882 | authority.split(':').next().unwrap_or(authority) |
| 883 | } |
| 884 | |
| 885 | /// Whether `host` is an IPv4/IPv6/name loopback address. |
| 886 | fn is_loopback_host(host: &str) -> bool { |
| 887 | let host = host.trim().trim_matches(|c| c == '[' || c == ']'); |
| 888 | if host.eq_ignore_ascii_case("localhost") { |
| 889 | return true; |
| 890 | } |
| 891 | // Parse real addresses rather than pattern-matching a `127.` prefix: the |
| 892 | // old `strip_prefix("127.") && 4 dot-parts` check classified |
| 893 | // `127.evil.example.com` as loopback (2026-08-04 review), which would let |
| 894 | // a hostile hostname inherit local-trust routing. `Ipv4Addr::is_loopback` |
| 895 | // is exactly the 127.0.0.0/8 block; `Ipv6Addr::is_loopback` is `::1`. |
| 896 | if let Ok(v4) = host.parse::<std::net::Ipv4Addr>() { |
| 897 | return v4.is_loopback(); |
| 898 | } |
| 899 | if let Ok(v6) = host.parse::<std::net::Ipv6Addr>() { |
| 900 | return v6.is_loopback(); |
| 901 | } |
| 902 | false |
| 903 | } |
| 904 | |
| 905 | #[cfg(test)] |
| 906 | mod loopback_tests { |
| 907 | use super::{endpoint_uses_insecure_http, is_loopback_host}; |
| 908 | |
| 909 | #[test] |
| 910 | fn loopback_matches_only_real_loopback_addresses() { |
| 911 | assert!(is_loopback_host("localhost")); |
| 912 | assert!(is_loopback_host("LocalHost")); |
| 913 | assert!(is_loopback_host("127.0.0.1")); |
| 914 | assert!(is_loopback_host("127.1.2.3")); // all of 127.0.0.0/8 |
| 915 | assert!(is_loopback_host("::1")); |
| 916 | assert!(is_loopback_host("[::1]")); |
| 917 | |
| 918 | // The 2026-08-04 regression: a hostile hostname that merely starts |
| 919 | // with `127.` and has four dot-parts must NOT be trusted as local. |
| 920 | assert!(!is_loopback_host("127.evil.example.com")); |
| 921 | assert!(!is_loopback_host("127.0.0.1.evil.com")); |
| 922 | assert!(!is_loopback_host("notlocalhost")); |
| 923 | assert!(!is_loopback_host("10.0.0.1")); |
| 924 | assert!(!is_loopback_host("localhost.evil.com")); |
| 925 | } |
| 926 | |
| 927 | #[test] |
| 928 | fn insecure_http_flags_a_hostile_127_lookalike() { |
| 929 | // loopback stays exempt (local runtimes use plain http) |
| 930 | assert!(!endpoint_uses_insecure_http("http://127.0.0.1:11434/v1")); |
| 931 | assert!(!endpoint_uses_insecure_http("http://localhost:8000/v1")); |
| 932 | // a real remote host dressed up as 127.* is insecure http |
| 933 | assert!(endpoint_uses_insecure_http( |
| 934 | "http://127.evil.example.com/v1" |
| 935 | )); |
| 936 | } |
| 937 | } |
| 938 |